With their research article titled “TripletCough: Cougher Identification and Verification from Contact-Free Smartphone-Based Audio Recordings Using Metric Learning”, the study team around Filipe Barata aim to solve an unresolved problem in cough detection. That is, to determine to whom a given cough belongs. This can be problematic in situations if a monitored patient shares a room with other people who cough. In such a scenario, the presence of other people’s coughs, so-called ambient coughs, leads to a very high and incorrect number of coughs being assigned to the patient. Therefore, algorithms are needed that are able to assign coughs to individuals. From a technical point of view, this brings the additional challenge of personalization, i.e., the monitoring system must have the ability to recognize the patient in addition to detecting coughs.

This study uses nocturnal cough recordings from asthma patients. The proposed approach to solving the personalization problem involves a triplet network architecture, which employs convolutional neural networks (CNNs). The CNNs of the triplet network learn an embedding function, which maps Mel spectrograms of cough recordings to an embedding space where they are more easily distinguishable. Using audio recordings of nocturnal coughs from asthmatic patients captured with a smartphone, the approach achieved a mean accuracy of 88% on identification tests and accuracy of 80% on verification tests. Furthermore, it outperformed human raters with regard to verification tests on average by 8% in accuracy.

For further details on this research article by Stefan Jokic, David Cleres, Frank Rassouli, Claudia Steurer-Stey, Milo A. Puhan, Martin Brutsche, Elgar Fleisch, and Filipe Barata, please refer to the full research article. The code and models are publicly available under: https://github.com/ADAMMA-CDHI-ETH-Zurich/TripletCough.

Reference:

Stefan Jokic, David Cleres, Frank Rassouli, Claudia Steurer-Stey, Milo A. Puhan, Martin Brutsche, Elgar Fleisch, and Filipe Barata. “TripletCough: Cougher Identification and Verification from Contact-Free Smartphone-Based Audio Recordings Using Metric Learning.” IEEE Journal of Biomedical and Health Informatics (2022). DOI: 10.1109/JBHI.2022.3152944.